Professional Documents
Culture Documents
Naqvi Et Al 2022 Decoding The Human Face Progress and Challenges in Understanding The Genetics of Craniofacial
Naqvi Et Al 2022 Decoding The Human Face Progress and Challenges in Understanding The Genetics of Craniofacial
Genetics of Craniofacial
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
Morphology
Sahin Naqvi,1,2,∗ Hanne Hoskens,3,4,∗ Franziska Wilke,5
Seth M. Weinberg,6,7,8 John R. Shaffer,6,7 Susan Walsh,5
Mark D. Shriver,9 Joanna Wysocka,1,10,11,†
and Peter Claes3,4,12,13,†
1
Department of Chemical and Systems Biology, Stanford University School of Medicine,
Stanford, California, USA; email: naqvi@stanford.edu, wysocka@stanford.edu
2
Department of Genetics, Stanford University School of Medicine, Stanford,
California, USA
3
Center for Processing Speech and Images, Department of Electrical Engineering, KU Leuven,
Leuven, Belgium; email: hanne.hoskens@kuleuven.be, peter.claes@kuleuven.be
4
Medical Imaging Research Center, University Hospitals Leuven, Leuven, Belgium
5
Department of Biology, Indiana University–Purdue University Indianapolis, Indianapolis,
Indiana, USA; email: fwilke@iu.edu, walshsus@iupui.edu
6
Department of Human Genetics, University of Pittsburgh, Pittsburgh, Pennsylvania, USA;
Annu. Rev. Genom. Hum. Genet. 2022. 23:383–412 email: smwst46@pitt.edu, john.r.shaffer@pitt.edu
7
Center for Craniofacial and Dental Genetics, Department of Oral and Craniofacial Sciences,
First published as a Review in Advance on
University of Pittsburgh, Pittsburgh, Pennsylvania, USA
April 28, 2022
8
Department of Anthropology, University of Pittsburgh, Pittsburgh, Pennsylvania, USA
The Annual Review of Genomics and Human Genetics 9
Department of Anthropology, The Pennsylvania State University, University Park,
is online at genom.annualreviews.org
Pennsylvania, USA; email: mds17@psu.edu
https://doi.org/10.1146/annurev-genom-120121- 10
Department of Developmental Biology, Stanford University School of Medicine, Stanford,
102607 California, USA
11
Copyright © 2022 by Annual Reviews. This work is Howard Hughes Medical Institute, Stanford University School of Medicine, Stanford,
licensed under a Creative Commons Attribution 4.0 California, USA
International License, which permits unrestricted 12
Department of Human Genetics, KU Leuven, Leuven, Belgium
use, distribution, and reproduction in any medium, 13
Murdoch Children’s Research Institute, Melbourne, Victoria, Australia
provided the original author and source are credited.
See credit lines of images or other third-party
material in this article for license information
∗
These authors contributed equally to this work
†
These authors jointly supervised this work
383
Keywords
craniofacial, development, genome-wide association study, GWAS, syndromes, gene regulation,
phenotyping
Abstract
Variations in the form of the human face, which plays a role in our individual identities and so-
cietal interactions, have fascinated scientists and artists alike. Here, we review our current under-
standing of the genetics underlying variation in craniofacial morphology and disease-associated
dysmorphology, synthesizing decades of progress on Mendelian syndromes in addition to more
recent results from genome-wide association studies of human facial shape and disease risk. We
also discuss the various approaches used to phenotype and quantify facial shape, which are of par-
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
ticular importance due to the complex, multipartite nature of the craniofacial form. We close by
discussing how experimental studies have contributed and will further contribute to our under-
standing of human genetic variation and then proposing future directions and applications for the
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
field.
INTRODUCTION
The human face and craniofacial structures exhibit a high degree of variation both among in-
dividuals of our own species and in comparison with those of other great apes (106). Changes
in human craniofacial structures that occurred since the divergence from our closest living evolu-
tionary relatives, chimpanzees and bonobos, facilitated the emergence of the uniquely human facial
appearance and may have contributed to adaptations associated with bipedal locomotion, diet (79),
enlarged brain size (114), and speech articulation (91). Consistent with the rapid evolution of hu-
man facial traits, modern human skulls display characteristic features that are distinct from those
of extinct hominins such as Homo erectus, archaic humans, and Neanderthals (5), whereas other
facial attributes (e.g., aspects of nose shape) have been postulated to mediate climate adaptation
in human populations (174).
In addition to facilitating biological adaptations, the human face is a strong target of sexual
selection and plays key roles in communication and other social interactions (53, 138). Cranio-
facial malformations—which are among the most common congenital diseases—therefore have
not only huge medical implications but also devastating social consequences for affected patients
and their families. Understanding the factors driving human craniofacial variation in health and
disease is essential for comprehending what makes our faces both human and individual and for
developing therapeutic strategies, treatment plans, and reconstruction techniques for a myriad of
craniofacial disorders.
In this review, we discuss the contributions of genetics to understanding morphological cranio-
facial variation. We begin with a brief review of human craniofacial development, including some
insights from studies of other species, before discussing studies that explore the overall heritability
of human facial shape. Since the face is a complex morphological structure, we first discuss various
phenotyping approaches and then provide an overview of craniofacial dysmorphism in rare syn-
dromes. We next discuss genome-wide association studies (GWASs) of facial shape and associated
insights into the genetic architecture of common variation, as well as the interplay between com-
mon and rare variation and associated conceptual models. Finally, we discuss post-GWAS analyses
in the form of fine mapping and experimental functional studies before describing future possible
directions for the field.
the first branchial arch gives rise to the paired maxillary and mandibular prominences, which at this
point are still separated by the stomodeum (the precursor of the mouth). The remaining branchial
arches give rise to noncraniofacial structures. In the fifth week, nasal placodes, from which the lat-
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
eral nasal prominence and medial nasal prominence are derived, appear as thickenings on each
side of the frontonasal prominence, and the mandibular prominences merge at the midline, con-
tributing to the mandible, chin, and lower lip. Fusion of the medial nasal prominences is followed
by fusion with the maxillary prominences approximately seven weeks after conception, leading to
the formation of the central structures of the nose (columella) and central upper lip (philtrum).
The sides and alae (wings) of the nose form when the lateral nasal prominences and the maxillary
prominences begin to merge. These existing structures grow and mature during the remaining
weeks of pregnancy (145).
The complex morphogenesis of the face, from the early specification and migration of CNCCs
to the development of the facial prominences and derived structures, requires tightly controlled
spatial and temporal signaling networks and gene expression patterns. Many of the mechanisms
underlying early specification and morphological events have been uncovered by decades of study
in model systems in which the early developmental transitions observed in human embryos are
conserved, such as mouse and chick embryos. While a comprehensive review of such studies is
beyond the scope of this work, two key concepts from this field emerge: first, that both cell-
autonomous patterning of neural crest cells and embryo region-specific environmental cues by
signaling ligands are key for proper craniofacial development, and second, that different combi-
nations of the same signaling pathways pattern specific craniofacial regions in distinct ways. A ma-
jor mechanism by which this patterning is achieved is by the combinatorial expression, through
both cell-autonomous and signaling-responsive mechanisms, of sequence-specific transcription
factors. The gene regulatory networks formed by these transcription factors have been delineated
and reviewed in detail by Simões-Costa & Bronner (142). The major signaling pathways mediating
environmental cues to CNCCs and other craniofacial cell types include fibroblast growth factor
(FGF) (115), bone morphogenetic protein (BMP) (50), Sonic hedgehog (Shh) (41), wingless/Int-1
(128), and transforming growth factor beta (TGF-β) (40), all of which have been reviewed in de-
tail elsewhere. Despite complex interactions between CNCCs and other cell types, such as the
epithelium, in the formation of the craniofacial structures, cross-species transplantation experi-
ments in birds [reviewed by Schneider (133)] have demonstrated that species-specific features of
the craniofacial complex are driven largely by the neural crest, suggesting that CNCCs may in
turn be the major factor determining variation in craniofacial morphology and risk for disease in
humans, as discussed in subsequent sections.
Postnatally, the craniofacial complex undergoes continuous growth and remodeling, with the
peak of growth occurring at puberty (145). The precise timing tends to be different for males and
Facial variation is determined largely by genetic variation in combination with diverse environ-
mental influences. The strong genetic component can be easily recognized and interpreted by per-
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
ceptual observation; for example, there is a strong resemblance of facial features across families,
and monozygotic twins are more alike than dizygotic twins. In addition, shared facial characteris-
tics within ancestries (81) and the sexes (84, 105) as well as the distinctive facial features associated
with particular genetic conditions (57) further demonstrate the extent to which genes influence
craniofacial form and appearance.
Craniofacial heritability studies have been performed to estimate the proportion of the phe-
notypic variation that can be explained by genetics, as compared with environmental effects. Tra-
ditionally, this is done by measuring the facial similarity between relatives, such as twins (39, 154,
162) or parents and offspring (60, 73). The widespread availability of whole-genome microar-
rays and recent advances in statistical methods allow for these traditional family-based designs
to be extended to large data sets of unrelated individuals where the degree of genetic relatedness
is determined from genome-wide single-nucleotide polymorphisms (SNPs) (28, 171). However,
these differences in study design, as well as differences in study populations and facial descrip-
tors employed, have produced widely varying estimates of heritability in the literature, making
direct comparisons across studies challenging. Research designs based on the similarity between
twins or siblings tend to overestimate heritability due to shared environmental sources of covari-
ance (common environment), whereas the lower bound is typically specified by the heritability
estimate obtained from common SNPs using approaches that assume only additive effects (44).
The acquisition of both genomic data and phenotypic measures for a sample of related individuals
poses an interesting avenue to dissect this bias. Furthermore, heritability is a dynamic rather than
a static descriptor, and estimates are expected to differ across populations and even within a single
population due to varying genetic and environmental influences (158).
Despite these differences, strong genetic influences have predominantly been found for cen-
tral midfacial parameters, with estimates exceeding 0.8 (i.e., >80% of variation is attributable to
genetics) (27, 60, 111, 154, 162). In particular, high heritability has consistently been reported for
aspects of nasal shape such as the position of the nasion, which is closely linked to the PAX3 locus
(1, 11, 19, 25, 61, 118, 164, 170). In addition to aspects of facial shape, facial size and allometry
have been found to be highly heritable (28, 39). By contrast, stronger environmental contributions
have been reported for the lower parts of the face, including the cheeks, mandible, and mouth,
which are known to be affected by nutrition (141), aging (119), and oral function (123). Other
nongenetic influences that can affect facial shape include prenatal exposure to teratogens [e.g.,
through maternal smoking (7) or alcohol intake (108, 150)], which may then lead to interactions
with genes involved in craniofacial development, and the effects of climate (174), geography (151),
and altitude (14) (e.g., through adaptation over evolutionary timescales).
predictable, so they can be controlled for statistically by means of regression in studies focused on
the genetics of craniofacial variation.
Various imaging systems have been used to capture the facial surface and/or underlying skele-
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
ton in two or three dimensions (136). Three-dimensional imaging techniques are preferable be-
cause they successfully preserve the multidimensional nature of facial morphology. In facial genetic
research today, approaches based on stereophotogrammetry and laser scanning are commonly
used to generate high-quality images of the three-dimensional facial surface. Different acquisition
systems are commercially available, all of which are comparable in terms of accuracy, with error
rates well under 1 mm, as reviewed elsewhere (120). However, caution is needed when combining
data obtained from different imaging systems into a single analysis in order to avoid systematic
bias (165). An important drawback of three-dimensional imaging systems is that they are more
expensive than their two-dimensional counterparts and are not always portable, although this is
now beginning to change. Two-dimensional descriptions (1, 11, 19) and even categorical scoring of
facial features (1, 121) remain valuable, especially in the context of large-scale data collection, but
with the introduction of the TrueDepth camera in the iPhone X, three-dimensional facial imaging
is becoming more accessible to the larger public (131). In medical imaging, cephalometry can be
used to assess facial structures in two dimensions, while devices such as computed tomography
and cone-beam computed tomography can preserve the three-dimensional structure of the un-
derlying craniofacial skeleton in addition to the facial surface (136). However, radiation exposure
generally limits their availability to clinical settings. By contrast, magnetic resonance imaging is a
noninvasive technique that is directly informative about facial soft tissue structures, and it has been
adopted in large-scale projects such as the UK Biobank (148) and the Adolescent Brain Cognitive
Development (ABCD) study (17). The disadvantages of magnetic resonance imaging are its high
cost and long acquisition times.
Landmarks can be placed on two- and three-dimensional representations of a face from which
various quantitative measurements or features can be derived. Early approaches did so manually,
focusing on well-defined anatomical points, but extensive effort has been put into the development
of automatic methods to obtain fast, accurate, and reproducible results [4, 36, 90; for an in-depth
overview of different landmarking approaches, we refer readers to a paper by Böhringer & de
Jong (10)]. In addition, automatic methods allow for sparse descriptions to be expanded toward
spatially dense configurations of quasi-landmarks covering the entire facial surface (26, 66). Here,
the locations of the quasi-landmarks do not have any biological meaning per se, but homology is
implied by each quasi-landmark that occupies the same position on the face relative to all other
quasi-landmarks. Landmark configurations can subsequently be represented as linear distances,
angles, and ratios measured between the landmarks (traditional anthropometrics), or the land-
mark x-y-z coordinates themselves can be analyzed [geometric morphometrics (175)]. The latter
using deep learning methods. Deep learning with neural networks is widely used in biomedical
image analysis (3), although it is fairly new to craniofacial genetic research, and most applications
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
still utilize two-dimensional data. For example, Face2Gene, a popular tool developed by FDNA
(Sunrise, Florida) to aid clinicians in the diagnosis of genetic disorders, applies deep learning algo-
rithms to two-dimensional photographs of patients to learn relevant facial features for syndrome
classification (54). Deep learning extensions for 3D surface or mesh data, referred to as geomet-
ric deep learning, are currently being developed (13, 101). However, the main challenge of deep
learning remains the large data burden. In addition, techniques are often tailored to solve specific
applications [e.g., facial classification (46)], and it is still unclear whether the learning outcomes
are also relevant biologically.
specific transcription factors, which are also in turn responsible for more cell-autonomous pattern-
ing of the craniofacial structures. General chromatin modifiers, on the other hand, are expected to
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
have critical roles in the maintenance of transcriptional programs and cell identity in virtually all
cells of the body, raising the question of why they are so prominent in craniofacial syndromes. One
explanation is that some of these general regulators have additional, specialized functions that are
particularly important for cells such as CNCCs (15); another is that CNCCs and their derivatives,
along with other cell types that contribute to the face, are uniquely sensitized to these general per-
turbations of the chromatin and transcription machinery (52). Which of these two explanations is
more common remains to be seen. Nevertheless, the presence of such coherent functional classes
of genes and the large overlap with loci discovered by common variant association studies as well
as mouse models (discussed in subsequent sections) illustrate the overall success of gene mapping
rare craniofacial syndromes.
390
ANKH BCOR ACAN CWC27 ANKRD11 ALX1 PAX3 BMP1 AKT1 ALDH18A1 COLEC10
ATP6V0A2 LZTR1 ANTXR1 PARN ARID1B ALX3 PAX6 BMP4 ARHGAP31 B3GALT6 COLEC11
CCBE1 PQBP1 ASXL1 ALX4 PBX1 BMPER BRAF B3GAT3 CPLANE1
Naqvi et al.
IPO8 MBD5
CDH1 CSF1R CHST3 DYNC1I2
KCNH1 RAI1 RBMX ATRX ARX PRRX1 CHUK
CDH2 CTNNB1 CHST14 EPB41L1
KCNJ2 SNIP1 SF3B4 BICRA BCL11B RAX EDN1 FGD1
CDH11 BPTF DLX3 RUNX2 CYP26B1 FLNA
KCNMA1 TCOF1 SMG9 EDNRA GNAS
COL1A1 DHCR7 FLNB
regulator
PAM16 ZMIZ1 SNRPN CHD7 DLX5 SATB2 EFNB1 HRAS
COL1A2 DPYSL5 IFT43
processing
EED ERF SIX3 ERBB3
RNA binding,
SLC10A7 ZMYM2 COL11A1 SPEN KRAS EBP LMNA
THOC6 EP300 FOXC1 SOX2 EVC2 MAPK1
SLC25A24 ZMYND11
Other transcriptional
DCHS1 FGFR1 ESCO2 MAPRE2
SLC35D1 FAM20C WBP11 EZH2 FOXL2 SOX9 FGFR2 MAP2K1 FTO MID1
TMCO1 FBN1 HDAC8 GATA6 TBX1 FGFR3 MAP2K2 GALNT2 NHS
FRAS1 CDC6 KAT5 GLI2 TBX2 FZD2 MTOR HCCS PCLO
ASPH FREM1 CDC45 KAT6B GLI3 TBX15 GDF5 PTPN11 HHAT RHOA
FBXL4 DPH1 GJA1 DDX11 KDM1A GSC TBX22 GH1 RAF1 HS2ST1 RTTN
GPC4 IL11RA HSD17B4 SPECC1L
Intracellular structure
RIPK4
replication
LRP4
DNA repair,
LTBP3 INPPL1 TNNI2
ITCH MGP ORC1 MORC2 MAF TFAP2B NOTCH2 RIT1
MAB21L2 TPM3
ADAMTS10 SUMO1 MMP14 ACTB NSD1 MEIS2 TP63 PTCH1 SOS2 NAA10 TRIO
modification
Extracellular matrix, cell adhesion
TRIM37 MYMK CPLANE1 PCGF2 MEOX1 TRPS1 PTCH2 ZPR1
ECE1 PTCH3 NAA15 WDPCP
KLHL7 UBE3B PKDCC DYNC1I2 PHF8 MITF TWIST1 PIGL
Other posttranslational
LONP1 PLOD3 EPB41L1 PHF21A MSX1 TWIST2 PIK3CA
PTPRF
RELN FLNA
Signaling ligand/receptor
Proteolysis
PEX1 IFT43 SMARCE1 NKX2-5 ZBTB20 SEMA3E ARCN1 PSAT1 IARS2
TASP1
PEX2 MAPRE2 SRCAP NKX2-6 ZEB2 SHH COG4 PTDSS1 RPL5
UBR1 SKI PYCR1
ZMPSTE24 PEX3 CENPF MID1 STAG2 OTX2 ZNF462 IFT172 RPL11
SMO SC5D
PEX5 CENPT NHS SOST PACS1 RPL15
TXNL4A
PEX6 CEP57 PCLO INTS1
Cytoskeleton
STRA6 SEC23A RPL35A
CCDC32 PEX10 CEP120 RTTN MED12 CDC6 SUFU SEC24D RPS7
MN1 PEX13 CKAP2L TAPT1 MED13L CDC45 TGFB3
Translation
Peroxisome
RSPRY1 PEX14 KIAA0753 TNNI2 POLR1A DDX11
chaperones
Chromosome
WNT3
machinery
Unknown
VPS35L NDN RPS24
replication
WNT5A
DNA repair,
Cell cycle
Intracellular trafficking,
regulation
ZSWIM6 PEX26 SMCHD1 WDPCP POLR3A ORC1 WNT10B WASHC5 NEK1 RPS26
organization, segregation
General transcription
Figure 1
Genes and pathways mutated in rare craniofacial disorders with Mendelian inheritance. The genes were identified by curating hits for the search term craniofacial from
the OMIM database, which were then organized into broad functional categories based partly on their PANTHER protein classes. Associations with GWASs were
defined by aggregating the candidate genes and loci from studies listed in Table 1. Genes that cause craniofacial phenotypes when mutated in mice (red) were found by
querying the Mouse Phenome Database. Blue shading indicates genes that GWASs have implicated in facial shape or nsCL/P. Abbreviations: GWAS, genome-wide
association study; nsCL/P, nonsyndromic cleft lip with or without cleft palate; OMIM, Online Mendelian Inheritance in Man; PANTHER, Protein Annotation Through
Evolutionary Relationship.
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
www.annualreviews.org
and PAX1
•
Shaffer et al. (137) 2016 Meta-analysis (N = 3,118) European Linear distances (N = 20) derived from Identified genome-wide significant
three-dimensional facial images associations at 7 loci; candidate
genes MAFB, PAX9, MIPOL1,
ALX3, HDAC8, and PAX1
Cole et al. (27) 2016 Discovery (N = 3,505), African Linear distances (N = 25), principal Identified and replicated genome-wide
replication (N = 2,390) components (N = 6), and facial size significant associations at 2 loci;
measures (N = 3) derived from candidate genes SCHIP1 and
three-dimensional facial images PDE8A
Lee et al. (88) 2017 Meta-analysis (N = 2,817) European Factors of linear distances (N = 23) Identified genome-wide significant
derived from three-dimensional facial associations at 3 loci; candidate
images genes PARK2 and FREM1
Crouch et al. (30) 2018 Discovery (N = 1,832), European Principal component extremes (N = 40) Identified genome-wide significant
replication (N = 1,567) derived from three-dimensional facial associations at 27 loci; replicated 3
images signals; candidate genes PCDH15,
MBTPS1, and TMEM163
391
segments (N = 63) derived from signals; candidate genes TBX15,
three-dimensional facial images ASPM, PKDCC, HOXD cluster,
PAX3, RAB7A/ACAD9, EPHB3/
DVL3, DCHS2, SUPT3H, RPS12/
EYA4, DLX6/DYNC1L1, BC039327,
SOX9, and KCTD15
(Continued)
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
Table 1 (Continued)
Study Year Study design Ancestry Phenotypes Key findings
Endo et al. (42) 2018 Meta-analysis (N = 11,311) East Asian Diverse skin-related traits (N = 7), Identified genome-wide significant
including the categorical scoring of associations at 1 locus affecting
double-edged eyelids double-edged eyelids; candidate
genes EMX2, and EMX2OS
Cha et al. (19) 2018 East Asian Linear distances, angles, areas, and Identified genome-wide significant
392
Discovery (N = 5,643),
replication (N = 1,926) curvature (N = 85) derived from associations at 5 loci; candidate
two-dimensional facial images genes OSR1-WDR35,
HOXD1-MTX2, WDR27, SOX9,
and DHX35
Howe et al. (62) 2018 Meta-analysis (N = 6,136) European Linear distance (N = 1) derived from Identified genome-wide significant
Naqvi et al.
three-dimensional facial images associations at 2 loci affecting
representing philtrum width philtrum width; candidate genes
HOXA1 and ENSG00000232633
Qiao et al. (125) 2018 Discovery (N = 1,709), European, East Linear distances (N = 10), principal Identified genome-wide significant
replication (N = 1,675) Asian, admixed components (N = 6), and partial associations at 6 loci; replicated 4
European– least-squares components (N = 6) signals; candidate genes COL23A1,
Asian derived from three-dimensional facial PCDH7, UBASH3B, and BMP2
images
Wu et al. (169) 2019 Discovery (N = 50) East Asian Linear distances (N = 48) derived from Identified genome-wide significant
three-dimensional computed associations at 4 loci; candidate
tomography scans of the skull genes RGPD3, IGSF3, SLC28A3,
and USP40
Xiong et al. (170) 2019 Discovery (N = 10,115), European, Latin Linear distances (N = 78) derived from Identified genome-wide significant
replication (N = 7,917) American, three-dimensional facial images associations at 24 loci; replicated 10
admixed signals; candidate genes CASZ1,
European– INTU, KIF6, TBX3, C14orf64,
Asian RPGRIP1L, PAX3, SFRP2, ROR2,
and PAX1
White et al. (164) 2021 Meta-analysis (N = 8,246) European Combined principal components Identified genome-wide significant
representing global-to-local facial associations at 203 loci and
segments (N = 63) derived from study-wide significant associations
three-dimensional facial images at 120 loci; found 117 new and 86
previously identified signals
Bonfante et al. (11) 2021 Discovery (N = 6,169) Latin American Linear distances, ratios, and angles Identified genome-wide significant
(N = 59) derived from associations at 32 loci; replicated 23
two-dimensional facial images signals and found 9 new signals
Indencleef et al. (67) 2021 Meta-analysis (N = 8,246) European Scores representing nsCL/P Identified genome-wide significant
endophenotype (N = 59) defined associations at 29 loci and
across global-to-local facial segments study-wide significant associations
(N = 63) derived from at 9 loci; 22 signals were previously
three-dimensional facial images identified in GWASs on normal-
range facial variation, and 18 signals
were near genes with strong
evidence in orofacial clefting
(Continued)
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
Table 1 (Continued)
Study Year Study design Ancestry Phenotypes Key findings
Huang et al. (65) 2021 Discovery (N = 2,659); East Asian Combined principal components Identified genome-wide significant
additionally tested representing different facial segments associations at 7 loci and study-wide
56 SNPs previously (N = 10) derived from significant associations at 4 loci;
reported in facial GWAS three-dimensional facial images replicated 24 SNPs from previously
literature reported genetic loci near DCHS2,
SUPT3H, HOXD1, SOX9, PAX3,
and EDAR; candidate genes
DENND1B, PISRT1, DCHS2/
SFRP2, and VPS13B
Liu et al. (94) 2021 Discovery (N = 2,329); European Combined principal components Identified significant associations at
tested low-frequency representing global-to-local facial 7 genes: HFE, NECTIN1, CARS2,
coding variants segments (N = 31) derived from LTB4R, TELO2, AR, and FTSJ1
(MAF < 1%) in 8,091 three-dimensional facial images
genes
www.annualreviews.org
Hoskens et al. (61) 2021 European Scores representing facial traits shared Identified genome-wide significant
•
Meta-analysis (N = 8,246)
by siblings (N = 1,048) defined across associations at 218 loci and study-
global-to-local facial segments wide significant associations at 38
(N = 63) derived from loci; found 109 new and 109
three-dimensional facial images previously identified signals
Liu et al. (93) 2021 Discovery (N = 2,595) African Combined principal components Identified genome-wide significant
representing global-to-local facial associations at 20 loci and
segments (N = 63) derived from study-wide significant associations
three-dimensional facial images at 6 loci; 10 signals were shared with
Europeans
Wang et al. (161) 2021 Meta-analysis (N = 9,674) East Asian Combined principal components Identified genome-wide significant
representing global-to-local facial associations at 244 loci and
segments (N = 63) derived from study-wide significant associations
three-dimensional facial images at 151 loci; found 130 new and 114
previously identified signals; 89
signals were shared with Europeans
393
Abbreviations: GWAS, genome-wide association study; MAF, minor allele frequency; nsCL/P, nonsyndromic cleft lip with or without cleft palate; SNP, single-nucleotide polymorphism.
found that 501 independent SNPs (r2 < 0.1), encompassing 303 loci, explained 13.7% of variance
of the full face, represented by the first 36 principal components as determined by parallel analysis.
In the same cohort, age, sex, and body mass index individually explained 7.0%, 12.2%, and 18.9%
of the variance, respectively. These results are consistent with a polygenic architecture typical of
complex traits, where many variants contribute individually small effects. As a comparison, for
human height (one of the most polygenic complex traits), 3,290 independent, genome-wide sig-
nificant variants explain 24.6% of variance (172). Despite having a polygenic architecture overall,
a small number of common variants can have quite large effects on facial shape; a recent study
specifically comparing the upper and lower tails of the distribution of facial shape (measured by
specific principal components describing shape variation) found three variants with frequencies
of approximately 10% and unusually large effects (odds ratios >4 for being in either tail) (30).
The two most common craniofacial birth defects, cleft lip with or without cleft palate (CL/P)
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
craniofacial syndromes, but approximately 70% (74) and 50% (140) of CL/P and craniosynostosis
cases, respectively, do not appear as part of a syndrome. Nonsyndromic CL/P (nsCL/P) occurs
in 1 of every 920 live births in the United States (117), while the prevalence of isolated cran-
iosynostosis has been estimated at between 1 in 2,100 and 1 in 2,500 live births (12, 87). There
have been numerous GWASs of nsCL/P (6, 8, 35, 51, 64, 89, 98–100, 102, 103, 107, 149); these
studies have been reviewed in detail elsewhere (152), but we also provide a combined list of vari-
ants and loci discovered by these studies in Supplemental Table 3. Together, these studies have
demonstrated a substantial contribution of common variants to nsCL/P risk, with one study find-
ing that approximately 30% of the variance in nsCL/P is attributable to all common SNPs (99).
Notably, the same study found that the top 24 loci explain 25% of nsCL/P risk (99) in European-
ancestry individuals, and another found that 26 loci explain approximately 11% of the variance
in a Chinese cohort (173), indicating substantially larger effect sizes for nsCL/P than for facial
shape variation. While nonsyndromic craniosynostosis common-variant GWASs have had less
statistical power for variant discovery due to their smaller sample sizes, variants at the BMP2 lo-
cus have been independently associated with two main subtypes, sagittal (76) and metopic (75)
craniosynostosis. Here, too, the effect sizes of lead variants are unusually large relative to both
facial shape variation and most common diseases (odds ratios >4 for sagittal and >2 for metopic
craniosynostosis).
While the vast majority of GWASs to date have sought to discover variants with additive ef-
fects on the phenotype of interest, genetic interactions—either with other variants (gene × gene
interactions, or epistasis) or with environmental variables—can be an additional source of vari-
ation. As a whole, GWASs of diverse phenotypes have found very few examples of large-effect
epistatic interactions (129), and facial shape GWASs are similar in that regard. Several general
explanations have been provided for this lack of epistasis, including a lack of statistical power to
detect epistatic effects (as they are far smaller than the main additive effects) and the testing of
tagging SNPs rather than the causal variant itself. Nevertheless, a few exceptions do exist. White
et al. (164) found four pairs of loci, each with its own additive associations with distinct multi-
variate facial phenotypes, which show significant epistasis. Furthermore, Liu et al. (95) described
an interaction between the PRICKLE1 and FOCAD loci for cranial base width underlying a vari-
ance quantitative trait locus (QTL) at PRICKLE1. With respect to craniofacial disease, three dis-
tinct modifiers of nsCL/P subtypes have been discovered in the last few years: the FAT4 locus
for cleft laterality (34), the PAX1 locus for bilaterality versus unilaterality (33), and a third locus
for cleft palate (16).
have revealed that many facial GWAS-associated loci contain genes implicated in rare cranio-
facial syndromes discussed above (Figure 2). For example, the first GWAS locus associated with
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
common facial variation was PAX3, rare variants in which are known to cause Waardenburg
syndrome. Additional candidate genes highlighted by GWASs include SOX9, GLI3, TWIST1,
ALX1/4, MSX1, TFAP2B, DLX5/6, FGFR2, and BMP2/4. Indeed, of the craniofacial syndrome
genes also highlighted by GWASs, there is a striking enrichment for sequence-specific transcrip-
tion factors (Figure 1). There are several possible explanations for such an enrichment, ranging
from the broad regulatory but cell type–specific roles of transcription factors to their inherent
dosage sensitivity. Such hypotheses would be best addressed with genome-wide mapping of tran-
scription factor targets as well as experimental measurements of their dosage sensitivities in the
relevant cell types, which are yet to be done systematically.
Minor allele
Major allele
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 X
Figure 2
Genomic locations and regional effects of genome-wide significant loci for facial shape. In the ideogram, the
genomic locations of lead SNPs identified through GWASs of facial shape in healthy individuals (listed in
Supplemental Table 2) are shown in blue, and genes implicated in rare craniofacial disorders with
Mendelian inheritance (listed in Figure 1) are shown in orange. Associated phenotypic effects of the PAX3
(lead SNP rs1370926), TWIST1 (lead SNP rs212672), ALX1 (lead SNP rs11609649), and SOX9 (lead SNP
rs9908442) loci are illustrated by the facial morphs, exaggerated in the direction of the minor and major
allele SNP variant based on the results of White et al. (164). Heat maps represent the normal displacement
(displacement in the direction locally normal to the facial surface) in each quasi-landmark going from the
minor to the major allele SNP variant. Blue indicates inward depression; red indicates outward protrusion.
Abbreviations: GWAS, genome-wide association study; SNP, single-nucleotide polymorphism.
normal-range facial shape variation with respect to effect sizes, many loci are associated with both
phenotypes. Studies explicitly testing the effects of nsCL/P risk loci on facial shape have found
multiple loci with highly significant associations on both (9, 68). Indeed, one study found a sig-
nificant association between a polygenic risk score for nsCL/P and the width of the philtrum,
disruption of which leads to nsCL/P (62). Another recent GWAS of healthy, unaffected individu-
als, which used endophenotypes (intermediate phenotypes that can be used as markers for disease
risk) derived from directly comparing facial scans of unaffected relatives of patients with nsCL/P
with those of controls, found both overlap with the results of previous facial shape and nsCL/P
Furthermore, this interaction occurred in the same pathway: BMP ligands bind and induce their
cognate receptors to phosphorylate SMAD and activate transcription of osteogenic genes, a pro-
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
Density of individuals
Disease prevalence
0.3
Craniosynostosis
Stickler
0.2 Saethre– Pierre
Chotzen Robin
Treacher
0.1 Collins
Crouzon
0.0
Cleft lip and palate liability Low High
Polygenicity
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
Genetics,
environment Cleft lip and palate liability
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
Philtrum width
Treacher
Collins Stickler
Pierre
2
Genetics,
environment
ime
Saethre–
pe d
Chozen
Sha
Craniosynostosis
Crouzon
Shape dimension 1
Figure 3
A multivariate model linking craniofacial shape and disease. (a) In standard liability threshold models, disease risk, which can be
modulated by genetics or the environment, manifests itself as a univariate trait, or endophenotype (e.g., philtrum width for
nonsyndromic cleft lip and palate). (b) Craniofacial disorders have a range of genetic architectures, from highly polygenic (right) to
monogenic with largely Mendelian inheritance (left). (c) In the proposed multivariate shape space model, variation along multiple axes
of shape (only two of which are shown for visualization purposes) leads to distinct yet overlapping zones of disease in the
multidimensional shape space. The craniofacial syndromes are arranged approximately according to phenotypic similarity. Facial
morphs for syndromes were created by first creating a univariate shape score distinguishing individuals with the syndrome from healthy
controls and then moving four standard deviations along the univariate syndromic axis.
sequence, Treacher Collins syndrome, and Stickler syndrome for nsCL/P and Saethre–Chotzen
and Crouzon syndromes for craniosynostosis). Such a model can also accommodate the variable
expressivity observed in many Mendelian syndromes, since additional factors, such as an individ-
ual’s polygenic background or developmental environment, could shift them away from the center
of the shape space zone corresponding to that syndrome.
ation studies such as GWASs inherently highlight regions of the genome and sets of linked SNPs
rather than exact causal variants underlying a trait. Numerous approaches, collectively termed fine
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
mapping, have been used to further narrow down candidate causal variants from GWASs. These
include cross-ancestry GWASs, where distinct patterns of linkage disequilibrium between popu-
lations can narrow down candidate variants (assuming the same causal variant underlies variation
in both populations); targeted fine-mapping GWASs, with denser panels of variants that allow
for imputation of smaller linkage blocks; and purely computational approaches. These statistical
approaches have been reviewed in detail by Schaid et al. (132). Here, we focus on fine-mapping
approaches based on experimental data.
Experimental fine-mapping approaches vary in both the strength of evidence provided and
scalability. Perhaps the most straightforward is to generate an epigenomic map from a cell type or
tissue thought to have an important role in the trait of interest, with the rationale that associated
variants located in regulatory regions are active in the cell type of interest. This approach has
been widely used by several GWASs of face shape and/or disease (25, 99, 163, 164), due in part
to its relative ease: Once epigenomic maps have been generated from sufficient cell numbers in
the relevant ex vivo– or in vitro–derived cell type, they can be reused in future GWASs, and the fine
mapping can be performed genome-wide. Stronger evidence can be provided by perturbing the
activity of candidate regulatory elements harboring associated variants and assessing the impact on
the expression of nearby genes. Yet stronger evidence results from re-creating the precise (typically
single-nucleotide) changes of each trait-associated variant and assessing their impact on regulatory
element activity and target gene expression.
Several studies have used experimental fine-mapping approaches beyond epigenomic maps.
Xiong et al. (170) carried out a GWAS of univariate facial distances from multiple ancestries and
performed both epigenomic fine mapping and luciferase reporter assays of resulting candidate
SNPs in neural crest cells, finding significant effects for four out of five SNPs tested. Two recent
in-depth studies of noncoding mutations underlying craniofacial syndromes provide further exam-
ples of what can be accomplished. Long et al. (97) combined epigenomic maps, enhancer reporter
assays, and endogenous deletion of enhancers in both in vitro–derived CNCCs and mice to show
that structural variants causing Pierre Robin sequence disrupt enhancers more than 1 Mb away
from SOX9. Hirsch et al. (59) used similar approaches to demonstrate that structural variants en-
compassing parts of HDAC9, a neighboring gene to the craniofacial transcription factor TWIST1,
cause craniosynostosis through disruption of TWIST1 enhancers. These studies were in part suc-
cessful at fine mapping the regulatory elements underlying disease-causing mutations because of
their relatively large effect size, but this becomes more challenging for variants associated with
common variation in facial shape or risk for nsCL/P or craniosynostosis.
variants to genes is a key component of understanding phenotypic variation. As with most GWASs
of complex traits and common diseases, most of the lead hits from studies of common facial shape
variation and disease lie in noncoding regions of the genome. Again, the gold standard for con-
necting variants to target genes is re-creating naturally occurring variants endogenously and as-
sessing their impact on nearby target genes, but as with fine mapping, higher-throughput ap-
proaches are necessary for mapping the target genes of the hundreds of variants associated with
facial shape or disease to date. One possible higher-throughput approach is the generation of
molecular QTL data sets, which involve measuring molecular traits such as RNA or protein levels
or splicing in tissue or cells from a population of individuals. Genetic variants in this population
can then be associated with molecular effects on a specific gene, and if such QTLs overlap with
trait-associated variants, the associated genes are candidates for mediating the variant–trait as-
sociation. Although eQTL data sets have yielded some success in fine mapping and target gene
identification for other traits, eQTLs can miss the relevant target genes for a number of reasons,
such as assaying cell types that are not important for the trait or insufficient statistical power
[reviewed by Umans et al. (155)]. Due to the limited availability of primary craniofacial tissue
from the relevant time points (i.e., during embryonic development), a more attainable avenue may
be to use induced pluripotent stem cells (iPSCs) from a population of individuals and differentiate
them into the relevant cell types (i.e., CNCCs or their chondrocyte/osteoblast derivatives). Studies
in iPSC-derived cardiomyocytes and neurons suggest that sample sizes of at least 20–80 individ-
uals are required to detect eQTLs with moderately large effect sizes when performing parallel
differentiations of iPSCs from different individuals combined with bulk RNA sequencing (134,
146). Recent innovative approaches based on pooled differentiations of iPSCs from many indi-
viduals, combined with single-cell RNA sequencing to assign cells to individuals based on genetic
polymorphisms, can both increase power by reducing noise due to differentiation batch effects
and allow for larger sample sizes to be assayed in fewer experiments (32, 71).
A complementary approach to population-based studies is to map the target gene of the reg-
ulatory element (i.e., the enhancer) in which candidate causal variants (identified by some fine-
mapping approach) reside. Broadly, these methods are based on the idea that enhancers come
into close physical proximity to the promoters of their target genes, a feature that can be mapped
genome-wide by chromosome conformation capture assays. Recent work has integrated such as-
says with other epigenomic data sets (i.e., measures of chromatin accessibility and histone modi-
fications) to predict target genes of enhancers (47), with appreciable precision in applications to
GWASs of common diseases, such as inflammatory bowel disease (113). Pairing chromosome con-
formation capture with chromatin immunoprecipitation (HiChIP) of active enhancer marks such
types include, for example, facial ectoderm, which secretes morphogens such as Shh, important for
patterning CNCCs during early development (41), as well as cell types important for later bone
remodeling, such as osteoclasts and recently identified skeletal stem cells (21). As discussed in the
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
introduction, cross-species chimeras suggest that CNCCs are the main drivers of species-specific
craniofacial patterning, whereas surface ectoderm is more important for the species-generic pat-
tern (133); such logic may also apply to within-species variation, but this remains to be seen. Ide-
ally, cell types would be assayed at a range of developmental times from primary human samples,
but this is challenging due to both the complex nature of such samples (i.e., containing many cell
types) and their limited availability.
The first challenge can be addressed using single-cell epigenomic profiling technologies, which
currently provide single-cell, genome-wide data at low depth, but when grouped by cell type pro-
vide depth similar to that of bulk assays (139). The overall lack of availability of primary sam-
ples (as well as the inability to use them for further functional tests, for obvious ethical reasons)
is one that may be partially ameliorated by the increasing success of in vitro organoid model-
ing approaches to recapitulate aspects of early human development. For example, a recent study
used micropatterned chips to achieve robust morphogenesis through folding and closure of the
neural tube, starting from commonly used pluripotent stem cells (77). Remarkably, these in vitro–
derived neural tubes gave rise to migratory neural crest cells, which could be expanded and profiled
further.
One potential solution to the lack of noncoding sequence conservation between human and
mouse would be to insert variants of the human regulatory element underlying a GWAS hit (deter-
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
mined using the experimental fine-mapping approaches discussed above) into the syntenic region
of the mouse genome and then assess the resulting effects on target gene expression and cranio-
facial morphology. However, it is possible, if not likely, that the regulatory landscape surrounding
the mouse orthologous locus has diverged enough that the perturbation may not reflect regulatory
changes equivalent to those influenced by the variant in the context of human development.
FUTURE DIRECTIONS
The human face is a highly variable and multipartite structure resulting from coordinated action
between diverse genetic and nongenetic factors. Recent advances in phenotyping strategies and
analytical approaches have substantially improved our understanding of the genetic contributions
to craniofacial morphology, and we expect this trend to continue in future years, though some
challenges remain.
GWASs have illuminated the polygenic nature of craniofacial morphology, uncovering more
than 300 genetic loci associated with one or multiple aspects of facial form. One of the major
challenges continues to be the identification of the associated variants for function and the mech-
anisms by which they alter gene function or expression. In addition, individual variants contribute
only a small proportion of the phenotypic variance, and a large number of genetic loci are yet to
be discovered, since they likely have effect sizes that are too small or allele frequencies that are too
rare to be detected with statistical significance using current sample sizes, which are still modest by
modern GWAS standards. Furthermore, epistatic effects between combinations of variants acting
on overlapping areas of the face shape space likely also have relatively small effects and thus remain
undiscovered. A natural way to increase statistical power is to increase the size of the study cohort,
for instance, through joint collaborative efforts. For traditional anthropometric measurements,
this is relatively straightforward, but it becomes more challenging when principal components or
multivariate descriptions of the face are used. The latter requires sharing individual raw images or
the underlying principal component analysis constructs, which is often not possible due to strict
privacy considerations.
Combining data sets is further complicated when individuals of diverse ancestral backgrounds
or developmental stages are included. Although the need for increasing sample diversity in genetic
studies is now widely appreciated, most facial shape GWASs have been performed in relatively ho-
mogeneous cohorts of primarily European descent. In cases of population heterogeneity, proper
statistical considerations are required to control for population stratification. One approach is to
stratify samples into major population groups, as determined by principal component analysis, and
craniofacial disorders, such an approach could be used to test whether distinct syndromes have
correlated but not identical common variant burdens, as predicted by the multivariate liability
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
threshold model. Large cohorts of healthy individuals would allow for the identification of in-
dividuals who have rare mutations in important craniofacial genes but no disease, which could
reveal additional environmental or genetic factors that reduce disease penetrance. Larger sample
sizes would also allow for the explicit modeling of gene–environment interactions, which remain
underexplored with respect to facial variation.
As our knowledge of the genetic architecture of the human face grows, it will become possi-
ble to use this information in concrete, real-life applications. Perhaps one of the most appealing
applications in human facial genetics is the prediction or reconstruction of facial features from bio-
logical material, for example, collected from patients [e.g., to guide orthodontic treatment or facial
reconstructive surgery (72)], recovered from skeletal remains [e.g., to visualize the appearance of
hominin ancestors (49)], or left at a crime scene [e.g., to narrow down potential suspects (23)].
Forensic DNA phenotyping, or the prediction of physical appearance from DNA, shows promise
when comparative DNA profiling fails (i.e., when no match between the unknown sample and
suspect or known profiles in a DNA database can be obtained) (82). The simultaneous prediction
of eye, hair, and skin color has already been validated forensically (20). However, current predic-
tion results for facial morphological traits do not reach high enough accuracy, and commercial
tools that claim to do so lack any scientific validation (166). Furthermore, previous studies have
shown that predictions are driven only by ancestry and sex (24, 92, 135), and complex patterns of
interactions among genes and/or with the environment further complicate matters (55). Methods
to predict facial appearance and infer DNA-related features have also raised concerns regarding
privacy and ethics (78, 83, 156), although the actual risk of reidentification was recently shown to
be smaller than previously claimed (157). The establishment of adequate legislation and regulatory
frameworks for any practice or application will be a necessity for its implementation.
DISCLOSURE STATEMENT
J.W. serves on the scientific boards of CAMP4, Paratus, the Whitehead Institute of the Mas-
sachusetts Institute of Technology, the Searle Scholars Program, and the International Institute
of Molecular and Cell Biology in Warsaw.
ACKNOWLEDGMENTS
J.W. was supported by the Howard Hughes Medical Institute, a Lorry Lokey endowed professor-
ship, and a Stinehart Reed award. S.N. was supported by a Helen Hay Whitney Fellowship. H.H.
and P.C. were supported by the National Institutes of Health (grant R01-DE027023), the KU
LITERATURE CITED
1. Adhikari K, Fuentes-Guajardo M, Quinto-Sánchez M, Mendoza-Revilla J, Chacón-Duque JC, et al.
2016. A genome-wide association scan implicates DCHS2, RUNX2, GLI3, PAX1 and EDAR in human
facial variation. Nat. Commun. 7:11616
2. Atkinson EG, Maihofer AX, Kanai M, Martin AR, Karczewski KJ, et al. 2021. Tractor uses local ancestry
to enable the inclusion of admixed individuals in GWAS and to boost power. Nat. Genet. 53:195–204
Annu. Rev. Genom. Hum. Genet. 2022.23:383-412. Downloaded from www.annualreviews.org
3. Baldi P. 2018. Deep learning in biomedical data science. Annu. Rev. Biomed. Data Sci. 1:181–205
4. Bannister JJ, Crites SR, Aponte JD, Katz DC, Wilms M, et al. 2020. Fully automatic landmarking of
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
28. Cole JB, Manyama M, Larson JR, Liberton DK, Ferrara TM, et al. 2017. Human facial shape and size
heritability and genetic correlations. Genetics 205:967–78
29. Cordero DR, Brugmann S, Chu Y, Bajpai R, Jame M, et al. 2011. Cranial neural crest cells on the move:
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
53. Guo J, Tan J, Yang Y, Zhou H, Hu S, et al. 2014. Variation and signatures of selection on the human face.
J. Hum. Evol. 75:143–52
54. Gurovich Y, Hanani Y, Bar O, Nadav G, Fleischer N, et al. 2019. Identifying facial phenotypes of genetic
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
77. Karzbrun E, Khankhel AH, Megale HC, Glasauer SMK, Wyle Y, et al. 2021. Human neural tube
morphogenesis in vitro by geometric constraints. Nature 599:268–72
78. Katsanis SH, Claes P, Doerr M, Cook-Deegan R, Tenenbaum JD, et al. 2021. A survey of U.S. public
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
perspectives on facial recognition technology and facial imaging data practices in health and research
contexts. PLOS ONE 16:e0257923
79. Katz DC, Grote MN, Weaver TD. 2017. Changes in human skull morphology across the agricultural
transition are consistent with softer diets in preindustrial farming groups. PNAS 114:9050–55
80. Kau CH, Richmond S. 2008. Three-dimensional analysis of facial morphology surface changes in
untreated children from 12 to 14 years of age. Am. J. Orthod. Dentofac. Orthop. 134:751–60
81. Kau CH, Richmond S, Zhurov A, Ovsenik M, Tawfik W, et al. 2010. Use of 3-dimensional surface acqui-
sition to study facial morphology in 5 populations. Am. J. Orthod. Dentofac. Orthop. 137(Suppl.):S56.e1–9
82. Kayser M. 2015. Forensic DNA phenotyping: predicting human appearance from crime scene material
for investigative purposes. Forensic Sci. Int. Genet. 18:33–48
83. Kayser M, de Knijff P. 2011. Improving human forensics through advances in genetics, genomics and
molecular biology. Nat. Rev. Genet. 12:179–92
84. Kesterke MJ, Raffensperger ZD, Heike CL, Cunningham ML, Hecht JT, et al. 2016. Using the 3D
Facial Norms Database to investigate craniofacial sexual dimorphism in healthy children, adolescents,
and adults. Biol. Sex Differ. 7:23
85. Knol MJ, Pawlak MA, Lamballais S, Terzikhan N, Hofer E, et al. 2022. Genetic architecture of orbital
telorism. Hum. Mol. Genet. 31:1531–43
86. Kohn LAP. 1991. The role of genetics in craniofacial morphology and growth. Annu. Rev. Anthropol.
20:261–78
87. Lajeunie E, Le Merrer M, Bonaïti-Pellie C, Marchac D, Renier D. 1995. Genetic study of nonsyndromic
coronal craniosynostosis. Am. J. Med. Genet. 55:500–4
88. Lee MK, Shaffer JR, Leslie EJ, Orlova E, Carlson JC, et al. 2017. Genome-wide association study of
facial morphology reveals novel associations with FREM1 and PARK2. PLOS ONE 12:e0176566
89. Leslie EJ, Carlson JC, Shaffer JR, Butali A, Buxó CJ, et al. 2017. Genome-wide meta-analyses of non-
syndromic orofacial clefts identify novel associations between FOXE1 and all orofacial clefts, and TP63
and cleft lip with or without cleft palate. Hum. Genet. 136:275–86
90. Li M, Cole JB, Manyama M, Larson JR, Liberton DK, et al. 2017. Rapid automated landmarking for
morphometric analysis of three-dimensional facial scans. J. Anat. 230:607–18
91. Lieberman P. 2007. The evolution of human speech: its anatomical and neural bases. Curr. Anthropol.
48:39–66
92. Lippert C, Sabatini R, Maher MC, Kang EY, Lee S, et al. 2017. Identification of individuals by trait
prediction using whole-genome sequencing data. PNAS 114:10166–71
93. Liu C, Lee MK, Naqvi S, Hoskens H, Liu D, et al. 2021. Genome scans of facial features in East Africans
and cross-population comparisons reveal novel associations. PLOS Genet. 17:e1009695
94. Liu D, Alhazmi N, Matthews H, Lee MK, Li J, et al. 2021. Impact of low-frequency coding variants on
human facial shape. Sci. Rep. 11:748
nonsyndromic cleft lip with or without cleft palate identify six new risk loci. Nat. Genet. 44:968–71
101. Mahdi SS, Nauwelaers N, Joris P, Bouritsas G, Gong S, et al. 2022. Matching 3D facial shape to
demographic properties by geometric metric learning: a part-based approach. IEEE Trans. Biometr.
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
123. Proffit W, Fields H, Larson B, Sarver D. 2018. Contemporary Orthodontics. Philadelphia: Elsevier. 6th ed.
124. Qian Y, Xiong Z, Li Y, Zhou H, Kayser M, et al. 2021. Knocking-out the human face genes TBX15 and
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
PAX1 in mice alters facial and other physical morphology. bioRxiv 2021.05.26.445773. https://doi.org/
10.1101/2021.05.26.445773
125. Qiao L, Yang Y, Fu P, Hu S, Zhou H, et al. 2018. Genome-wide variants of Eurasian facial shape differ-
entiation and a prospective model of DNA based face prediction. J. Genet. Genom. 45:419–32
126. Ransom RC, Carter AC, Salhotra A, Leavitt T, Marecic O, et al. 2018. Mechanoresponsive stem cells
acquire neural crest fate in jaw regeneration. Nature 563:514–21
127. Reardon W, Winter RM, Rutland P, Pulleyn LJ, Jones BM, et al. 1994. Mutations in the fibroblast
growth factor receptor 2 gene cause Crouzon syndrome. Nat. Genet. 8:98–103
128. Reynolds K, Kumari P, Sepulveda Rincon L, Gu R, Ji Y, et al. 2019. Wnt signaling in orofacial clefts:
crosstalk, pathogenesis and models. Dis. Models Mech. 12:dmm037051
129. Ritchie MD, Van Steen K. 2018. The search for gene-gene interactions in genome-wide association
studies: challenges in abundance of methods, practical considerations, and biological interpretation. Ann.
Transl. Med. 6:157
130. Roosenboom J, Indencleef K, Lee MK, Hoskens H, White JD, et al. 2018. SNPs associated with
testosterone levels influence human facial morphology. Front. Genet. 9:497
131. Rudy HL, Wake N, Yee J, Garfein ES, Tepper OM. 2020. Three-dimensional facial scanning at the fin-
gertips of patients and surgeons: accuracy and precision testing of iPhone X three-dimensional scanner.
Plast. Reconstr. Surg. 146:1407–17
132. Schaid DJ, Chen W, Larson NB. 2018. From genome-wide associations to candidate causal variants by
statistical fine-mapping. Nat. Rev. Genet. 19:491–504
133. Schneider RA. 2018. Neural crest and the origin of species-specific pattern. Genesis 56:e23219
134. Schwartzentruber J, Foskolou S, Kilpinen H, Rodrigues J, Alasoo K, et al. 2018. Molecular and functional
variation in iPSC-derived sensory neurons. Nat. Genet. 50:54–61
135. Sero D, Zaidi A, Li J, White JD, Zarzar TBG, et al. 2019. Facial recognition from DNA using face-to-
DNA classifiers. Nat. Commun. 10:2557
136. Sforza C, de Menezes M, Ferrario V. 2013. Soft- and hard-tissue facial anthropometry in three dimen-
sions: what’s new. J. Anthropol. Sci. 91:159–84
137. Shaffer JR, Orlova E, Lee MK, Leslie EJ, Raffensperger ZD, et al. 2016. Genome-wide association study
reveals multiple loci influencing normal human facial morphology. PLOS Genet. 12:e1006149
138. Sheehan MJ, Nachman MW. 2014. Morphological and population genomic evidence that human faces
have evolved to signal individual identity. Nat. Commun. 5:4800
139. Shema E, Bernstein BE, Buenrostro JD. 2019. Single-cell and single-molecule epigenomics to uncover
genome regulation at unprecedented resolution. Nat. Genet. 51:19–25
140. Shin JH, Persing JA. 2007. Nonsyndromic craniosynostosis and deformational plagiocephaly. In Grabb
and Smith’s Plastic Surgery, ed. CH Thorne, RW Beasley, SJ Aston, SP Bartlett, GC Gurtner, SL Spear,
pp. 226–36. Philadelphia: Lippincott Williams & Wilkins. 6th ed.
172. Yengo L, Sidorenko J, Kemper KE, Zheng Z, Wood AR, et al. 2018. Meta-analysis of genome-wide
association studies for height and body mass index in ∼700 000 individuals of European ancestry. Hum.
Mol. Genet. 27:3641–49
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
173. Yu Y, Zuo X, He M, Gao J, Fu Y, et al. 2017. Genome-wide analyses of non-syndromic cleft lip with
palate identify 14 novel loci and genetic heterogeneity. Nat. Commun. 8:14364
174. Zaidi AA, Mattern BC, Claes P, McEcoy B, Hughes C, et al. 2017. Investigating the case of human nose
shape and climate adaptation. PLOS Genet. 13:e1006616
175. Zelditch ML, Swiderski DL, Sheets HD, Fink WL. 2004. Geometric Morphometrics for Biologists: A Primer.
Boston: Academic
Annual Review of
Genomics and
Contents Human Genetics
Errata
Access provided by 84.76.54.3 on 03/13/24. See copyright for approved use.
An online log of corrections to Annual Review of Genomics and Human Genetics articles
may be found at http://www.annualreviews.org/errata/genom